File size: 15,408 Bytes
632066d | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 | # Major TOM — Index
The Major TOM Index is a global metadata catalog for the [Major TOM](https://github.com/ESA-PhiLab/Major-TOM) grid at 10 km resolution. It provides a single entry point to discover, filter, and select tiles across sensors, locations, and time without downloading any imagery.
The index covers over 5 million tiles spanning the entire Earth. Each tile corresponds to a 1056 × 1056 px patch (10.56 × 10.56 km) aligned to Sentinel-2 MGRS tiles at 10 m resolution. Every tile is enriched with terrain, climate, soil, socioeconomic, and administrative attributes derived from public Earth Engine datasets.
**What can you do with this index?**
- **Find tiles by location.** Filter by country, state, MGRS tile code, or bounding box using the GeoParquet geometry column.
- **Select tiles by environmental criteria.** Want arid, high-elevation tiles? Filter by `climate:precipitation < 200` and `terrain:elevation > 3000`.
- **Stratify sampling for training sets.** Use the enrichment columns to build geographically and environmentally balanced splits for foundation model pretraining.
- **Link to imagery.** The `land_s2` and `land_l8` files include sensor-specific image IDs (`s2:id_gee`, `l8:id_gee`) that point directly to the source products in Google Earth Engine.
- **Use the ELLIOT splits.** The `elliot.parquet` file provides pre-built monotemporal and temporal splits designed for multi-sensor, multi-temporal EO research.
All files are self-contained GeoParquet with ZSTD compression, sorted by `majortom:code_1000km` → `majortom:code_100km` → `id` for efficient spatial predicate pushdown.
## Schema
Columns are organized into namespaces. Each namespace groups related attributes.
### Grid (`majortom:`)
Tile identity and spatial reference within the Major TOM grid system.
| Column | Type | Description |
|--------|------|-------------|
| `id` | string | Unique tile identifier (e.g. `MT10_770U_395R`). |
| `majortom:code_100km` | string | Parent 100 km grid cell. Used for spatial grouping. |
| `majortom:code_1000km` | string | Parent 1000 km grid cell. Used for coarse-level partitioning. |
| `majortom:crs` | string | Native UTM CRS of the tile (e.g. `EPSG:32647`). |
| `majortom:mgrs_tile` | string | MGRS tile code (e.g. `47WNS`). Links to Sentinel-2 tiling grid. |
| `majortom:mgrs_n` | uint8 | Number of overlapping MGRS tiles (1 after deduplication). |
| `majortom:mgrs_candidates` | list\<string\> | All candidate MGRS tiles before deduplication. |
| `majortom:footprint_pct` | float | Percentage of tile covered by the assigned MGRS tile. |
| `majortom:geotransform` | list\<int32\> | Snapped affine geotransform [originX, scaleX, shearX, originY, shearY, scaleY]. |
| `majortom:geotransform_raw` | list\<double\> | Original (unsnapped) affine geotransform. |
### STAC (`stac:`)
Spatial and temporal reference following STAC conventions. Present in `land_s2` and `land_l8` only, where it replaces the `majortom:` grid columns.
| Column | Type | Description |
|--------|------|-------------|
| `stac:crs` | string | Coordinate reference system. |
| `stac:geotransform` | list\<int64\> | Affine geotransform for the image patch. |
| `stac:tensor_shape` | list\<int32\> | Shape of the image tensor [bands, height, width]. |
| `stac:time_start` | int64 | Acquisition start time (Unix timestamp). |
| `stac:time_end` | int64 | Acquisition end time (Unix timestamp). |
### Sentinel-2 (`s2:`)
Sensor metadata for the assigned Sentinel-2 image. Present in `land_s2` only.
| Column | Type | Description |
|--------|------|-------------|
| `s2:id_gee` | string | Google Earth Engine image ID. Use this to fetch the actual imagery. |
| `s2:product_id` | string | ESA product identifier. |
| `s2:spacecraft` | string | Spacecraft name (Sentinel-2A or Sentinel-2B). |
| `s2:processing_baseline` | string | Processing baseline version. |
| `s2:orbit_number` | uint16 | Relative orbit number. |
| `s2:mean_solar_azimuth` | float | Mean solar azimuth angle, averaged across all bands and detectors (degrees). |
| `s2:mean_solar_zenith` | float | Mean solar zenith angle, averaged across all bands and detectors (degrees). |
| `s2:mean_view_azimuth` | float | Mean viewing azimuth angle from band B8 (degrees). |
| `s2:mean_view_zenith` | float | Mean viewing zenith angle from band B8 (degrees). |
| `s2:reflectance_conversion` | float | Reflectance conversion factor (U correction). |
> **Note on solar vs viewing angles.** The sun has a single position relative to the scene, so ESA provides one solar azimuth and one solar zenith averaged across all bands. Viewing angles are different: Sentinel-2 uses a pushbroom sensor where each spectral band has its own detector array in the focal plane, each observing from a slightly different angle. That is why GEE provides per-band viewing angles (`MEAN_INCIDENCE_*_ANGLE_B1` through `_B12`). We use band B8 (NIR, 10 m) as the reference because it is at native 10 m resolution and sits near the center of the focal plane, making it a representative proxy for the viewing geometry of the 10 m and 20 m bands.
### Landsat 8/9 (`l8:`)
Sensor metadata for the assigned Landsat image. Present in `land_l8` only.
| Column | Type | Description |
|--------|------|-------------|
| `l8:id_gee` | string | Google Earth Engine image ID. Use this to fetch the actual imagery. |
| `l8:product_id` | string | USGS product identifier. |
| `l8:spacecraft` | string | Spacecraft name (Landsat 8 or Landsat 9). |
| `l8:collection_number` | uint8 | USGS Collection number. |
| `l8:collection_category` | string | Collection category (T1, T2, RT). |
| `l8:processing_software` | string | Processing software version. |
| `l8:wrs_path` | uint16 | WRS-2 path number. |
| `l8:wrs_row` | uint16 | WRS-2 row number. |
| `l8:cloud_cover` | float | Scene cloud cover percentage. |
| `l8:sun_azimuth` | float | Sun azimuth angle (degrees). |
| `l8:sun_elevation` | float | Sun elevation angle (degrees). |
| `l8:earth_sun_distance` | float | Earth-Sun distance (astronomical units). |
| `l8:image_quality_oli` | uint8 | OLI image quality score. |
| `l8:roll_angle` | float | Spacecraft roll angle (degrees). |
### Terrain (`terrain:`)
| Column | Type | Range | Description |
|--------|------|-------|-------------|
| `terrain:elevation` | float | ~-420 to 8,849 (m) | Mean elevation in meters from the [Copernicus GLO-30 DEM](https://gee-community-catalog.org/projects/glo30/), a 30 m resolution Digital Surface Model derived from TanDEM-X radar satellite data (2011 to 2015). Includes buildings, infrastructure, and vegetation. Uses the EGM2008 vertical datum. |
### Climate (`climate:`)
| Column | Type | Range | Description |
|--------|------|-------|-------------|
| `climate:precipitation` | float | 0+ (mm/year) | Mean annual precipitation estimated from [GPM](https://gpm.nasa.gov/) (Global Precipitation Measurement) satellite data, aggregated as a long-term annual mean. |
| `climate:temperature` | float | ~-40 to 50 (°C) | Mean annual land surface temperature estimated from [MODIS LST](https://developers.google.com/earth-engine/datasets/catalog/MODIS_061_MOD11A1) satellite data, aggregated as a long-term annual mean. |
### Soil (`soil:`)
Surface-layer soil properties from the [OpenLandMap](https://openlandmap.org/) dataset, derived from machine learning predictions on global soil survey data at 250 m resolution.
| Column | Type | Range | Description |
|--------|------|-------|-------------|
| `soil:clay` | float | 0 to 100 (%) | Clay content weight fraction at 0 cm depth. [Source](https://developers.google.com/earth-engine/datasets/catalog/OpenLandMap_SOL_SOL_CLAY-WFRACTION_USDA-3A1A1A_M_v02). |
| `soil:sand` | float | 0 to 100 (%) | Sand content weight fraction at 0 cm depth. [Source](https://developers.google.com/earth-engine/datasets/catalog/OpenLandMap_SOL_SOL_SAND-WFRACTION_USDA-3A1A1A_M_v02). |
| `soil:carbon` | float | 0+ (g/kg) | Soil organic carbon content at 0 cm depth. [Source](https://developers.google.com/earth-engine/datasets/catalog/OpenLandMap_SOL_SOL_ORGANIC-CARBON_USDA-6A1C_M_v02). |
| `soil:bulk_density` | float | 0+ (kg/m³) | Fine-earth bulk density at 0 cm depth. [Source](https://developers.google.com/earth-engine/datasets/catalog/OpenLandMap_SOL_SOL_BULKDENS-FINEEARTH_USDA-4A1H_M_v02). |
| `soil:ph` | float | ~3 to 10 | Soil pH in water at 0 cm depth. [Source](https://developers.google.com/earth-engine/datasets/catalog/OpenLandMap_SOL_SOL_PH-H2O_USDA-4C1A2A_M_v02). |
### Socioeconomic (`socio:`)
| Column | Type | Range | Description |
|--------|------|-------|-------------|
| `socio:gdp` | float | 0+ (USD) | GDP per capita at purchasing power parity (PPP, constant 2021 USD) for the year 2022. From the [Kummu et al. (2025)](https://doi.org/10.1038/s41597-025-04487-x) gridded dataset, downscaled to admin-2 level (43,501 units) at 5 arc-min resolution. [GEE catalog](https://gee-community-catalog.org/projects/gridded_gdp_hdi/). |
| `socio:population` | float | 0+ (people) | Estimated number of people per grid cell from the Meta [High Resolution Settlement Layer](https://dataforgood.facebook.com/dfg/tools/high-resolution-population-density-maps) (HRSL). Uses satellite imagery and census data at ~30 m resolution. |
| `socio:human_modification` | float | 0.0 to 1.0 | Cumulative degree of human modification of terrestrial ecosystems from the [Global Human Modification v3](https://doi.org/10.1038/s41597-025-04892-2) (Theobald et al. 2025). Combines the spatial footprint and intensity of 13 stressors across five categories: settlement, agriculture, transportation, mining/energy, and electrical infrastructure. 0 = no modification, 1 = fully modified. 300 m resolution. [GEE catalog](https://gee-community-catalog.org/projects/ghm-v3/). |
| `socio:cisi` | float | 0.0 to 1.0 | [Critical Infrastructure Spatial Index](https://doi.org/10.1038/s41597-022-01218-4) (Nirandjan et al. 2022). Aggregates OpenStreetMap data on 39 types of critical infrastructure across seven systems: transportation, energy, telecommunication, waste, water, education, and health. 0 = no infrastructure, 1 = highest density. 0.10° resolution. [GEE catalog](https://gee-community-catalog.org/projects/cisi/). |
### Administrative (`admin:`)
Human-readable administrative boundary names resolved from rasterized boundary datasets.
| Column | Type | Description |
|--------|------|-------------|
| `admin:country` | string | Country name. Tiles over ocean/lakes are labeled `Ocean/Sea/Lakes`. |
| `admin:state` | string | State or province name. |
| `admin:district` | string | District or county name. |
### Other
| Column | Type | Description |
|--------|------|-------------|
| `geometry` | binary (WKB) | Tile geometry. All files include GeoParquet metadata for spatial queries. |
| `split` | string | ELLIOT split assignment: `monotemporal` or `temporal`. Present in `elliot.parquet` only. |
## Files
| File | Rows | Columns | Size | Description |
|------|-----:|--------:|-----:|-------------|
| `global.parquet` | 5,055,204 | 26 | 146 MB | Every 10 km tile on Earth. The complete grid with all enrichment columns. |
| `land.parquet` | 2,767,104 | 26 | 91 MB | Tiles covered by land-observing sensors (Sentinel-2 and Landsat). Same schema as global. |
| `land_s2.parquet` | 2,547,253 | 34 | 127 MB | Land tiles with a Sentinel-2 image assigned. Adds `stac:` and `s2:` sensor metadata. |
| `land_l8.parquet` | 2,255,537 | 38 | 97 MB | Land tiles with a Landsat 8/9 image assigned. Adds `stac:` and `l8:` sensor metadata. |
| `elliot.parquet` | 279,166 | 27 | 14 MB | ELLIOT subset with monotemporal and temporal split assignments. Same enrichment as global plus `split` column. |
### Namespace availability per file
| Namespace | global | land | land_s2 | land_l8 | elliot |
|-----------|:------:|:----:|:-------:|:-------:|:------:|
| `majortom:` | ✓ | ✓ | | | ✓ |
| `stac:` | | | ✓ | ✓ | |
| `s2:` | | | ✓ | | |
| `l8:` | | | | ✓ | |
| `terrain:` | ✓ | ✓ | ✓ | ✓ | ✓ |
| `climate:` | ✓ | ✓ | ✓ | ✓ | ✓ |
| `soil:` | ✓ | ✓ | ✓ | ✓ | ✓ |
| `socio:` | ✓ | ✓ | ✓ | ✓ | ✓ |
| `admin:` | ✓ | ✓ | ✓ | ✓ | ✓ |
| `split` | | | | | ✓ |
| `geometry` | ✓ | ✓ | ✓ | ✓ | ✓ |
## Quick Start
### DuckDB
```sql
INSTALL spatial;
LOAD spatial;
-- Count tiles per country in South America
SELECT "admin:country", COUNT(*) as n_tiles
FROM 'https://data.source.coop/majortom/index/land_s2.parquet'
WHERE "admin:country" IN ('Peru', 'Brazil', 'Colombia', 'Chile', 'Argentina')
GROUP BY "admin:country"
ORDER BY n_tiles DESC;
-- Find high-elevation, arid Sentinel-2 tiles
SELECT id, "s2:id_gee", "terrain:elevation", "climate:precipitation"
FROM 'https://data.source.coop/majortom/index/land_s2.parquet'
WHERE "terrain:elevation" > 3000
AND "climate:precipitation" < 200
LIMIT 20;
```
### Pandas
```python
import pandas as pd
# Load land tiles with Sentinel-2 metadata
url = "https://data.source.coop/majortom/index/land_s2.parquet"
df = pd.read_parquet(url)
# Filter by country
peru = df[df["admin:country"] == "Peru"]
print(f"Peru: {len(peru):,} tiles")
# Get ELLIOT splits
elliot = pd.read_parquet(
"https://data.source.coop/majortom/index/elliot.parquet"
)
print(elliot["split"].value_counts())
```
## ELLIOT Splits
The `elliot.parquet` file contains 279,166 tiles selected for the [ELLIOT project](https://elliot-ai.eu/) multi-temporal dataset extension. Tile locations were sampled using hierarchical spherical k-means (530 × 528 = 279,840 clusters) over [AlphaEarth Foundation](https://source.coop/asterisklabs/alphaearth) embeddings to ensure global environmental diversity.
The `split` column defines two subsets:
- **Monotemporal** (250,000 tiles). One cloud-free image per sensor per location. Designed for tasks where spatial coverage matters more than temporal depth: land cover classification, feature extraction, or pretraining foundation models on diverse global scenes.
- **Temporal** (29,166 tiles). Multiple observations per location across time. Designed for tasks that require temporal context: change detection, phenology tracking, seasonal compositing, or training models that learn from multi-temporal sequences. This subset is further divided into monthly cadence (12,500 tiles × 12 timesteps) and five-daily cadence (16,666 tiles × 6 timesteps).
## License
This dataset is released under [CC-BY-4.0](https://creativecommons.org/licenses/by/4.0/).
## Citation
```bibtex
@inproceedings{Francis2024MajorTOM,
author = {Francis, Alistair and Czerkawski, Mikolaj},
title = {Major TOM: Expandable Datasets for Earth Observation},
booktitle = {IGARSS 2024 - IEEE International Geoscience and Remote Sensing Symposium},
year = {2024},
pages = {2935--2940},
doi = {10.1109/IGARSS53475.2024.10640760}
}
```
## Acknowledgments
This work was supported by the [ELLIOT project](https://elliot-ai.eu/), funded by the European Union under grant agreement No. 101214398. Views and opinions expressed are those of the authors only and do not necessarily reflect those of the European Union.
<p align="center">
<a href="images/elliot.png"><img src="images/asterisk.png" alt="ELLIOT" height="60"></a>
<a href="images/asterisk.png"><img src="images/elliot.png" alt="Asterisk Labs" height="60"></a>
</p> |